On 5 August, five teams inside the Rust project — compiler, libs, types, rustdoc, bootstrap and their subteams — ratified a policy governing large language models in the rust-lang/rust monorepo. It was authored by Jynn Nelson and published on the Inside Rust blog.
The line it draws
The policy's own summary is a list of permitted verbs: it is fine to use an LLM to answer questions, analyze, distill, refine, check, suggest, review — but not to create. Text originally produced by a model must not be submitted as issue descriptions, pull-request descriptions, GitHub comments, documentation, non-trivial source comments or compiler diagnostics.
What is still allowed
Machine translation, trivial changes, bug discovery the contributor has verified, and code changes that were pre-arranged with maintainers, non-critical, thoroughly tested and well reviewed. Everything public that came from a model must be marked as such. LLM-authored code is explicitly held to a stricter standard than human code, and reviewers are under no obligation to review it at all.
Why maintainers, not lawyers, wrote it
The stated rationale is about review economics. Polished prose and tidy diffs no longer signal that anyone did the work; review bandwidth is the project's scarcest resource; and pasting model output into a thread transfers effort from author to reviewer. The policy is a bandwidth defence, not a position on AI.
Narrower than the headlines
It is not a project-wide ban and does not cover Cargo, crates.io or the wider ecosystem. A separate project-wide RFC (rust-lang/rfcs #3959) is still open. What landed is one repository's house rule — which is exactly why it is enforceable.